def wrapped(logger):
with tempfile.TemporaryDirectory() as p:
path = os.path.join(p, 'download')
with logger.duration(f'downloading {url}'):
util.download(url, path)
with logger.duration('extracting vecs'):
with zipfile.ZipFile(path) as z:
zip_file_name = url.split('/')[-1][:-4] + ext
out_file_name = os.path.join(p, zip_file_name)
z.extract(zip_file_name, p)
with logger.duration(f'loading vecs into memory'):
terms = []
weights = []
with open(out_file_name, 'rt') as f:
for line in f:
cols = line.split()
if len(cols) == 2:
continue # First line includes # of terms and dim, we can ignore
if weights and len(weights[0]) != len(cols) - 1:
logger.warn(
f'problem parsing line, skipping {line[:20]}...'
)
else:
terms.append(cols[0])
weights.append([float(c) for c in cols[1:]])
weights = np.array(weights)
return terms, weights
def init(self, force=False):
base_path = util.path_dataset(self)
idxs = [self.index, self.index_stem, self.doc_store]
self._init_indices_parallel(idxs, self._init_iter_collection(), force)
qrels_file = os.path.join(base_path, 'qrels.robust2004.txt')
if (force or not os.path.exists(qrels_file)) and self._confirm_dua():
util.download(**_FILES['qrels'], file_name=qrels_file)
for fold in FOLDS:
fold_qrels_file = os.path.join(base_path, f'{fold}.qrels')
if (force or not os.path.exists(fold_qrels_file)):
all_qrels = trec.read_qrels_dict(qrels_file)
fold_qrels = {
qid: dids
for qid, dids in all_qrels.items() if qid in FOLDS[fold]
}
trec.write_qrels_dict(fold_qrels_file, fold_qrels)
query_file = os.path.join(base_path, 'topics.txt')
if (force or not os.path.exists(query_file)) and self._confirm_dua():
query_file_stream = util.download_stream(**_FILES['queries'],
encoding='utf8')
with util.finialized_file(query_file, 'wt') as f:
plaintext.write_tsv(f,
trec.parse_query_format(query_file_stream))
def init(self, force=False):
idxs = [self.index, self.index_stem, self.doc_store]
self._init_indices_parallel(idxs, self._init_iter_collection(), force)
train_qrels = os.path.join(util.path_dataset(self), 'train.qrels.txt')
valid_qrels = os.path.join(util.path_dataset(self), 'valid.qrels.txt')
if (force or not os.path.exists(train_qrels)
or not os.path.exists(valid_qrels)) and self._confirm_dua():
source_stream = util.download_stream(
'https://ciir.cs.umass.edu/downloads/Antique/antique-train.qrel',
encoding='utf8')
with util.finialized_file(train_qrels, 'wt') as tf, \
util.finialized_file(valid_qrels, 'wt') as vf:
for line in source_stream:
cols = line.strip().split()
if cols[0] in VALIDATION_QIDS:
vf.write(' '.join(cols) + '\n')
else:
tf.write(' '.join(cols) + '\n')
train_queries = os.path.join(util.path_dataset(self),
'train.queries.txt')
valid_queries = os.path.join(util.path_dataset(self),
'valid.queries.txt')
if (force or not os.path.exists(train_queries)
or not os.path.exists(valid_queries)) and self._confirm_dua():
source_stream = util.download_stream(
'https://ciir.cs.umass.edu/downloads/Antique/antique-train-queries.txt',
encoding='utf8')
train, valid = [], []
for cols in plaintext.read_tsv(source_stream):
if cols[0] in VALIDATION_QIDS:
valid.append(cols)
else:
train.append(cols)
plaintext.write_tsv(train_queries, train)
plaintext.write_tsv(valid_queries, valid)
test_qrels = os.path.join(util.path_dataset(self), 'test.qrels.txt')
if (force or not os.path.exists(test_qrels)) and self._confirm_dua():
util.download(
'https://ciir.cs.umass.edu/downloads/Antique/antique-test.qrel',
test_qrels)
test_queries = os.path.join(util.path_dataset(self),
'test.queries.txt')
if (force or not os.path.exists(test_queries)) and self._confirm_dua():
util.download(
'https://ciir.cs.umass.edu/downloads/Antique/antique-test-queries.txt',
test_queries)
def wrapped(logger):
with tempfile.TemporaryDirectory() as p:
vocab_path = os.path.join(p, 'vocab')
with logger.duration(f'downloading {url}'):
util.download(url, vocab_path)
with logger.duration(f'loading binary {vocab_path}'):
vectors = KeyedVectors.load_word2vec_format(vocab_path,
binary=True)
vocab_path += '.txt'
with logger.duration(f'saving text {vocab_path}'):
vectors.save_word2vec_format(vocab_path)
with logger.duration(f'reading embedding'):
weights = None
terms = []
for i, values in enumerate(
plaintext.read_sv(vocab_path, sep=' ')):
if i == 0:
weights = np.ndarray((int(values[0]), int(values[1])))
else:
term, values = values[0], values[1:]
terms.append(term)
weights[i - 1] = [float(v) for v in values]
return terms, np.array(weights)
def wrapped(logger, get_kernels=False):
with tempfile.TemporaryDirectory() as p:
if not get_kernels:
vocab_path = os.path.join(p, 'vocab')
with logger.duration(f'downloading {base_url}vocab'):
util.download(base_url + 'vocab', vocab_path)
with logger.duration(f'reading vocab'):
v = {}
for term, idx in plaintext.read_tsv(vocab_path):
v[int(idx)] = term
terms = [None] * (max(v.keys()) + 1)
for idx, term in v.items():
terms[idx] = term
embedding_path = os.path.join(p, 'embedding')
with logger.duration(f'downloading {base_url}embedding'):
util.download(base_url + 'embedding', embedding_path)
with logger.duration(f'reading embedding'):
weights = None
for values in plaintext.read_sv(embedding_path, sep=' '):
if len(values) == 2:
weights = np.ndarray(
(int(values[0]), int(values[1])))
else:
idx, values = values[0], values[1:]
weights[int(idx)] = [float(v) for v in values]
return terms, weights
else: # get_kernels
w, b = [], []
for f in range(1, 4):
url = f'{base_url}filter{f}'
path = os.path.join(p, f'filter{f}')
with logger.duration(f'downloading {url}'):
util.download(url, path)
with logger.duration(f'reading filter{f}'):
weights, biases = None, None
for i, values in enumerate(
plaintext.read_sv(path, sep=' ')):
if i == 0:
weights = np.ndarray(
(int(values[0]) * int(values[1]),
int(values[2])))
elif i == 1:
biases = np.array([float(v) for v in values])
else:
weights[:, i -
2] = [float(v) for v in values if v]
weights = weights.reshape(f, -1, weights.shape[1])
weights = np.transpose(weights, (2, 1, 0))
w.append(weights)
b.append(biases)
return w, b
def init(self, force=False):
needs_docs = []
for index in [self.index_stem, self.index_stem_2020, self.doc_store]:
if force or not index.built():
needs_docs.append(index)
if needs_docs and self._confirm_dua():
with contextlib.ExitStack() as stack:
doc_iter = self._init_iter_collection()
doc_iter = self.logger.pbar(doc_iter, desc='articles')
doc_iters = util.blocking_tee(doc_iter, len(needs_docs))
for idx, it in zip(needs_docs, doc_iters):
if idx is self.index_stem_2020:
it = (d for d in it if '2020' in d.data['date'])
stack.enter_context(
util.CtxtThread(functools.partial(idx.build, it)))
path = os.path.join(util.path_dataset(self), 'rnd1.tsv')
if not os.path.exists(path) and self._confirm_dua():
with util.download_tmp('https://ir.nist.gov/covidSubmit/data/topics-rnd1.xml', expected_md5="cf1b605222f45f7dbc90ca8e4d9b2c31") as f, \
util.finialized_file(path, 'wt') as fout:
soup = BeautifulSoup(f.read(), 'lxml-xml')
for topic in soup.find_all('topic'):
qid = topic['number']
plaintext.write_tsv(fout, [
(qid, 'query', topic.find('query').get_text()),
(qid, 'quest', topic.find('question').get_text()),
(qid, 'narr', topic.find('narrative').get_text()),
])
udel_flag = path + '.includes_udel'
if not os.path.exists(udel_flag):
with open(path,
'at') as fout, util.finialized_file(udel_flag, 'wt'):
with util.download_tmp(
'https://raw.githubusercontent.com/castorini/anserini/master/src/main/resources/topics-and-qrels/topics.covid-round1-udel.xml',
expected_md5="2915cf59ae222f0aa20b2a671f67fd7a") as f:
soup = BeautifulSoup(f.read(), 'lxml-xml')
for topic in soup.find_all('topic'):
qid = topic['number']
plaintext.write_tsv(fout, [
(qid, 'udel', topic.find('query').get_text()),
])
path = os.path.join(util.path_dataset(self), 'rnd2.tsv')
if not os.path.exists(path) and self._confirm_dua():
with util.download_tmp('https://ir.nist.gov/covidSubmit/data/topics-rnd2.xml', expected_md5="550129e71c83de3fb4d6d29a172c5842") as f, \
util.finialized_file(path, 'wt') as fout:
soup = BeautifulSoup(f.read(), 'lxml-xml')
for topic in soup.find_all('topic'):
qid = topic['number']
plaintext.write_tsv(fout, [
(qid, 'query', topic.find('query').get_text()),
(qid, 'quest', topic.find('question').get_text()),
(qid, 'narr', topic.find('narrative').get_text()),
])
udel_flag = path + '.includes_udel'
if not os.path.exists(udel_flag):
with open(path,
'at') as fout, util.finialized_file(udel_flag, 'wt'):
with util.download_tmp(
'https://raw.githubusercontent.com/castorini/anserini/master/src/main/resources/topics-and-qrels/topics.covid-round2-udel.xml',
expected_md5="a8988734e6f812921d5125249c197985") as f:
soup = BeautifulSoup(f.read(), 'lxml-xml')
for topic in soup.find_all('topic'):
qid = topic['number']
plaintext.write_tsv(fout, [
(qid, 'udel', topic.find('query').get_text()),
])
path = os.path.join(util.path_dataset(self), 'rnd1.qrels')
if not os.path.exists(path) and self._confirm_dua():
util.download(
'https://ir.nist.gov/covidSubmit/data/qrels-rnd1.txt',
path,
expected_md5="d58586df5823e7d1d0b3619a73b31518")
def init(self, force=False):
idxs = [self.index_stem, self.doc_store]
self._init_indices_parallel(idxs, self._init_iter_collection(), force)
base_path = util.path_dataset(self)
needs_queries = []
if force or not os.path.exists(
os.path.join(base_path, 'train.queries.tsv')):
needs_queries.append(lambda it: plaintext.write_tsv(
os.path.join(base_path, 'train.queries.tsv'),
((qid, txt) for file, qid, txt in it
if file == 'queries.train.tsv' and qid not in MINI_DEV)))
if force or not os.path.exists(
os.path.join(base_path, 'minidev.queries.tsv')):
needs_queries.append(lambda it: plaintext.write_tsv(
os.path.join(base_path, 'minidev.queries.tsv'),
((qid, txt) for file, qid, txt in it
if file == 'queries.train.tsv' and qid in MINI_DEV)))
if force or not os.path.exists(
os.path.join(base_path, 'dev.queries.tsv')):
needs_queries.append(lambda it: plaintext.write_tsv(
os.path.join(base_path, 'dev.queries.tsv'),
((qid, txt) for file, qid, txt in it
if file == 'queries.dev.tsv')))
if force or not os.path.exists(
os.path.join(base_path, 'eval.queries.tsv')):
needs_queries.append(lambda it: plaintext.write_tsv(
os.path.join(base_path, 'eval.queries.tsv'),
((qid, txt) for file, qid, txt in it
if file == 'queries.eval.tsv')))
if needs_queries and self._confirm_dua():
with util.download_tmp(_SOURCES['queries']) as f, \
tarfile.open(fileobj=f) as tarf, \
contextlib.ExitStack() as ctxt:
def _extr_subf(subf):
for qid, txt in plaintext.read_tsv(
io.TextIOWrapper(tarf.extractfile(subf))):
yield subf, qid, txt
query_iter = [
_extr_subf('queries.train.tsv'),
_extr_subf('queries.dev.tsv'),
_extr_subf('queries.eval.tsv')
]
query_iter = tqdm(itertools.chain(*query_iter), desc='queries')
query_iters = util.blocking_tee(query_iter, len(needs_queries))
for fn, it in zip(needs_queries, query_iters):
ctxt.enter_context(
util.CtxtThread(functools.partial(fn, it)))
file = os.path.join(base_path, 'train.qrels')
if (force or not os.path.exists(file)) and self._confirm_dua():
stream = util.download_stream(_SOURCES['train-qrels'], 'utf8')
with util.finialized_file(file, 'wt') as out:
for qid, _, did, score in plaintext.read_tsv(stream):
if qid not in MINI_DEV:
trec.write_qrels(out, [(qid, did, score)])
file = os.path.join(base_path, 'minidev.qrels')
if (force or not os.path.exists(file)) and self._confirm_dua():
stream = util.download_stream(_SOURCES['train-qrels'], 'utf8')
with util.finialized_file(file, 'wt') as out:
for qid, _, did, score in plaintext.read_tsv(stream):
if qid in MINI_DEV:
trec.write_qrels(out, [(qid, did, score)])
file = os.path.join(base_path, 'dev.qrels')
if (force or not os.path.exists(file)) and self._confirm_dua():
stream = util.download_stream(_SOURCES['dev-qrels'], 'utf8')
with util.finialized_file(file, 'wt') as out:
for qid, _, did, score in plaintext.read_tsv(stream):
trec.write_qrels(out, [(qid, did, score)])
file = os.path.join(base_path, 'train.mspairs.gz')
if not os.path.exists(file) and os.path.exists(
os.path.join(base_path, 'qidpidtriples.train.full')):
# legacy
os.rename(os.path.join(base_path, 'qidpidtriples.train.full'),
file)
if (force or not os.path.exists(file)) and self._confirm_dua():
util.download(_SOURCES['qidpidtriples.train.full'], file)
if not self.config['init_skip_msrun']:
for file_name, subf in [('dev.msrun', 'top1000.dev'),
('eval.msrun', 'top1000.eval'),
('train.msrun', 'top1000.train.txt')]:
file = os.path.join(base_path, file_name)
if (force or not os.path.exists(file)) and self._confirm_dua():
run = {}
with util.download_tmp(_SOURCES[file_name]) as f, \
tarfile.open(fileobj=f) as tarf:
for qid, did, _, _ in tqdm(
plaintext.read_tsv(
io.TextIOWrapper(tarf.extractfile(subf)))):
if qid not in run:
run[qid] = {}
run[qid][did] = 0.
if file_name == 'train.msrun':
minidev = {
qid: dids
for qid, dids in run.items() if qid in MINI_DEV
}
with self.logger.duration('writing minidev.msrun'):
trec.write_run_dict(
os.path.join(base_path, 'minidev.msrun'),
minidev)
run = {
qid: dids
for qid, dids in run.items() if qid not in MINI_DEV
}
with self.logger.duration(f'writing {file_name}'):
trec.write_run_dict(file, run)
query_path = os.path.join(base_path, 'trec2019.queries.tsv')
if (force or not os.path.exists(query_path)) and self._confirm_dua():
stream = util.download_stream(_SOURCES['trec2019.queries'], 'utf8')
plaintext.write_tsv(query_path, plaintext.read_tsv(stream))
msrun_path = os.path.join(base_path, 'trec2019.msrun')
if (force or not os.path.exists(msrun_path)) and self._confirm_dua():
run = {}
with util.download_stream(_SOURCES['trec2019.msrun'],
'utf8') as stream:
for qid, did, _, _ in plaintext.read_tsv(stream):
if qid not in run:
run[qid] = {}
run[qid][did] = 0.
with util.finialized_file(msrun_path, 'wt') as f:
trec.write_run_dict(f, run)
qrels_path = os.path.join(base_path, 'trec2019.qrels')
if not os.path.exists(qrels_path) and self._confirm_dua():
util.download(_SOURCES['trec2019.qrels'], qrels_path)
qrels_path = os.path.join(base_path, 'judgedtrec2019.qrels')
if not os.path.exists(qrels_path):
os.symlink('trec2019.qrels', qrels_path)
query_path = os.path.join(base_path, 'judgedtrec2019.queries.tsv')
judged_qids = util.Lazy(
lambda: trec.read_qrels_dict(qrels_path).keys())
if (force or not os.path.exists(query_path)):
with util.finialized_file(query_path, 'wt') as f:
for qid, qtext in plaintext.read_tsv(
os.path.join(base_path, 'trec2019.queries.tsv')):
if qid in judged_qids():
plaintext.write_tsv(f, [(qid, qtext)])
msrun_path = os.path.join(base_path, 'judgedtrec2019.msrun')
if (force or not os.path.exists(msrun_path)) and self._confirm_dua():
with util.finialized_file(msrun_path, 'wt') as f:
for qid, dids in trec.read_run_dict(
os.path.join(base_path, 'trec2019.msrun')).items():
if qid in judged_qids():
trec.write_run_dict(f, {qid: dids})
# A subset of dev that only contains queries that have relevance judgments
judgeddev_path = os.path.join(base_path, 'judgeddev')
judged_qids = util.Lazy(lambda: trec.read_qrels_dict(
os.path.join(base_path, 'dev.qrels')).keys())
if not os.path.exists(f'{judgeddev_path}.qrels'):
os.symlink('dev.qrels', f'{judgeddev_path}.qrels')
if not os.path.exists(f'{judgeddev_path}.queries.tsv'):
with util.finialized_file(f'{judgeddev_path}.queries.tsv',
'wt') as f:
for qid, qtext in plaintext.read_tsv(
os.path.join(base_path, 'dev.queries.tsv')):
if qid in judged_qids():
plaintext.write_tsv(f, [(qid, qtext)])
if self.config['init_skip_msrun']:
if not os.path.exists(f'{judgeddev_path}.msrun'):
with util.finialized_file(f'{judgeddev_path}.msrun',
'wt') as f:
for qid, dids in trec.read_run_dict(
os.path.join(base_path, 'dev.msrun')).items():
if qid in judged_qids():
trec.write_run_dict(f, {qid: dids})
if not self.config['init_skip_train10']:
file = os.path.join(base_path, 'train10.queries.tsv')
if not os.path.exists(file):
with util.finialized_file(file, 'wt') as fout:
for qid, qtext in self.logger.pbar(
plaintext.read_tsv(
os.path.join(base_path, 'train.queries.tsv')),
desc='filtering queries for train10'):
if int(qid) % 10 == 0:
plaintext.write_tsv(fout, [(qid, qtext)])
file = os.path.join(base_path, 'train10.qrels')
if not os.path.exists(file):
with util.finialized_file(file, 'wt') as fout, open(
os.path.join(base_path, 'train.qrels'), 'rt') as fin:
for line in self.logger.pbar(
fin, desc='filtering qrels for train10'):
qid = line.split()[0]
if int(qid) % 10 == 0:
fout.write(line)
if not self.config['init_skip_msrun']:
file = os.path.join(base_path, 'train10.msrun')
if not os.path.exists(file):
with util.finialized_file(file, 'wt') as fout, open(
os.path.join(base_path, 'train.msrun'),
'rt') as fin:
for line in self.logger.pbar(
fin, desc='filtering msrun for train10'):
qid = line.split()[0]
if int(qid) % 10 == 0:
fout.write(line)
file = os.path.join(base_path, 'train10.mspairs.gz')
if not os.path.exists(file):
with gzip.open(file, 'wt') as fout, gzip.open(
os.path.join(base_path, 'train.mspairs.gz'),
'rt') as fin:
for qid, did1, did2 in self.logger.pbar(
plaintext.read_tsv(fin),
desc='filtering mspairs for train10'):
if int(qid) % 10 == 0:
plaintext.write_tsv(fout, [(qid, did1, did2)])
if not self.config['init_skip_train_med']:
med_qids = util.Lazy(
lambda: {
qid.strip()
for qid in util.download_stream(
'https://raw.githubusercontent.com/Georgetown-IR-Lab/covid-neural-ir/master/med-msmarco-train.txt',
'utf8',
expected_md5="dc5199de7d4a872c361f89f08b1163ef")
})
file = os.path.join(base_path, 'train_med.queries.tsv')
if not os.path.exists(file):
with util.finialized_file(file, 'wt') as fout:
for qid, qtext in self.logger.pbar(
plaintext.read_tsv(
os.path.join(base_path, 'train.queries.tsv')),
desc='filtering queries for train_med'):
if qid in med_qids():
plaintext.write_tsv(fout, [(qid, qtext)])
file = os.path.join(base_path, 'train_med.qrels')
if not os.path.exists(file):
with util.finialized_file(file, 'wt') as fout, open(
os.path.join(base_path, 'train.qrels'), 'rt') as fin:
for line in self.logger.pbar(
fin, desc='filtering qrels for train_med'):
qid = line.split()[0]
if qid in med_qids():
fout.write(line)
if not self.config['init_skip_msrun']:
file = os.path.join(base_path, 'train_med.msrun')
if not os.path.exists(file):
with util.finialized_file(file, 'wt') as fout, open(
os.path.join(base_path, 'train.msrun'),
'rt') as fin:
for line in self.logger.pbar(
fin, desc='filtering msrun for train_med'):
qid = line.split()[0]
if qid in med_qids():
fout.write(line)
file = os.path.join(base_path, 'train_med.mspairs.gz')
if not os.path.exists(file):
with gzip.open(file, 'wt') as fout, gzip.open(
os.path.join(base_path, 'train.mspairs.gz'),
'rt') as fin:
for qid, did1, did2 in self.logger.pbar(
plaintext.read_tsv(fin),
desc='filtering mspairs for train_med'):
if qid in med_qids():
plaintext.write_tsv(fout, [(qid, did1, did2)])